* Update trainer and model flows to accommodate sparseml Disable FP16 on QAT start (huggingface#12) * Override LRScheduler when using LRModifiers * Disable FP16 on QAT start * keep wrapped scaler object for training after disabling Using QATMatMul in DistilBERT model class (huggingface#41) Removed double quantization of output of context layer. (huggingface#45) Fix DataParallel validation forward signatures (huggingface#47) * Fix: DataParallel validation forward signatures * Update: generalize forward_fn selection Best model after epoch (huggingface#46) fix sclaer check for non fp16 mode in trainer (huggingface#38) Mobilebert QAT (huggingface#55) * Remove duplicate quantization of vocabulary. enable a QATWrapper for non-parameterized matmuls in BERT self attention (huggingface#9) * Utils and auxillary changes update Zoo stub loading for SparseZoo 1.1 refactor (huggingface#54) add flag to signal NM integration is active (huggingface#32) Add recipe_name to file names * Fix errors introduced in manual cherry-pick upgrade Co-authored-by: Benjamin Fineran <bfineran@users.noreply.github.com>

truncated normal initializer #38

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions